experiment thus invites replication with otheroutgroups.The intervention’s durable effect on support

for a nondiscrimination law also suggests opti-mism for the public sphere. Over the past cen-tury, political campaigns have increasingly reliedonmass communication to reach voters (31). How-ever, facing difficulty persuading a polarizingpublic with these strategies (22), campaigns in-creasingly eschewmaking the case for their posi-tions and instead focus on rousing enthusiasm ofvoters who already agree with them (32). Theseshifts undermine basic aspirations for democraticdiscourse. However, these findings suggest that itmay be in campaigns’ own best interest to placerenewed emphasis on a personal exchange of ini-tially opposing views, even regarding controver-sial issues and across partisan lines.

REFERENCES AND NOTES

1. American Psychological Association, Declaration for the UNWorld Conference Against Racism, Racial Discrimination,Xenophobia, and Related Intolerance (2001; www.apa.org/pi/oema/programs/racism/apa-delegation-report.pdf).

2. R. D. Enos, Proc. Natl. Acad. Sci. U.S.A. 111, 3699–3704(2014).

3. T. F. Pettigrew, Annu. Rev. Sociol. 24, 77–103 (1998).4. E. L. Paluck, D. P. Green, Annu. Rev. Psychol. 60, 339–367

(2009).5. D. O. Sears, C. L. Funk, J. Polit. 61, 1–28 (1999).6. M. Tesler, Am. J. Polit. Sci. 59, 806–824 (2015).7. E. L. Paluck, J. Pers. Soc. Psychol. 96, 574–587 (2009).8. E. L. Paluck et al., PLOS ONE 10, e0138610 (2015).9. S. W. Cook, in Nebraska Symposium on Motivation,

W. J. Arnold, D. Levine, Eds. (Univ. of Nebraska Press, 1969),pp. 179–236.

10. C. V. Laar, S. Levin, S. Sinclair, J. Sidanius, J. Exp. Soc. Psychol.41, 329–345 (2005).

11. R. E. Petty, C. P. Haugtvedt, S. M. Smith, in Attitude Strength:Antecedents and Consequences, R. E. Petty, J. A. Krosnick, Eds.(Erlbaum Associates, Mahwah, NJ, 1995), pp. 93–130.

12. G. Ku, C. S. Wang, A. D. Galinsky, Res. Organ. Behav., 10.1016/j.riob.2015.07.003 (2015).

13. A. D. Galinsky, G. B. Moskowitz, J. Pers. Soc. Psychol. 78,708–724 (2000).

14. A. R. Flores, Polit. Groups Identities 3, 398–416(2015).

15. M. Fernandez, A. Blinder, Opponents of Houston rightsmeasure focused on bathrooms, and won, N.Y. Times (4 Nov.2015); available at www.nytimes.com/2015/11/05/us/houston-anti-discrimination-bathroom-ordinance.html(accessed 25 Jan. 2016).

16. H. Gehlbach, M. E. Brinkworth, Teach. Coll. Rec. 114, 226–254(2012).

17. D. W. Nickerson, T. Rogers, Psychol. Sci. 21, 194–199(2010).

18. E. J. Finkel, E. B. Slotter, L. B. Luchies, G. M. Walton, J. J. Gross,Psychol. Sci. 24, 1595–1601 (2013).

19. D. W. Nickerson, Polit. Anal. 13, 233–252 (2005).20. K. Sherrill, PS Polit. Sci. Polit. 29, 469–473

(1996).21. A. S. Yang, Public Opin. Q. 61, 477–507 (1997).22. A. S. Gerber, J. G. Gimpel, D. P. Green, D. R. Shaw, Am. Polit.

Sci. Rev. 105, 135–150 (2011).23. J. R. Zaller, The Nature and Origins of Mass Opinion (Cambridge

Univ. Press, 1992).24. D. Chong, J. N. Druckman, Annu. Rev. Polit. Sci. 10, 103–126

(2007).25. S. J. Hill, J. Lo, L. Vavreck, J. Zaller, Polit. Commun. 30,

521–547 (2013).26. A. R. Todd, A. D. Galinsky, Soc. Personality Psychol. Compass

8, 374–387 (2014).27. M. J. LaCour, D. P. Green, Science 346, 1366–1369

(2014).28. M. McNutt, Science 348, 1100 (2015).29. H. McGee, Personal route to reach public central to Yes

campaign, Irish Times (14 May 2015); available atwww.irishtimes.com/news/politics/personal-route-to-reach-

public-central-to-yes-campaign-1.2211282 (accessed25 Jan. 2016).

30. H. Han, How Organizations Develop Activists: CivicAssociations and Leadership in the 21st Century (OxfordUniv. Press, 2014).

31. T. Skocpol, Diminished Democracy: From Membership toManagement in American Civic Life (Univ. of Oklahoma Press,2003).

32. C. Panagopoulos, Party Polit. 22, 179–190(2015).

ACKNOWLEDGMENTS

The authors acknowledge funding from the Gill Foundation for thisresearch. D.B. acknowledges the National Science FoundationGraduate Research Fellowship Program for support whileconducting this research. Replication data and code are availableat http://dx.doi.org/10.7910/DVN/WKR39N. Neither of theauthors is affiliated with the Los Angeles LGBT Center or SAVE or

received compensation for this research. A preanalysis plan wasfiled after the data were collected but before they were analyzedwith the treatment indicator, and revised preanalysis plans werealso filed before the 3-week and 6-week surveys. These areavailable at the Evidence in Governance and Politics site (www.egap.org), ID 20150707AA. This research was approved by theUniversity of California–Berkeley Committee for the Protection ofHuman Subjects, Protocol ID 2015-04-7508.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/352/6282/220/suppl/DC1Materials and MethodsFig. S1Tables S1 to S25References (33–49)

30 November 2015; accepted 17 February 201610.1126/science.aad9713

CLIMATE CHANGE

Observational constraints onmixed-phase clouds imply higherclimate sensitivityIvy Tan,1* Trude Storelvmo,1 Mark D. Zelinka2

Global climate model (GCM) estimates of the equilibrium global mean surface temperatureresponse to a doubling of atmospheric CO2, measured by the equilibrium climate sensitivity(ECS), range from 2.0° to 4.6°C. Clouds are among the leading causes of this uncertainty.Here we show that the ECS can be up to 1.3°C higher in simulations where mixed-phase cloudsconsisting of ice crystals and supercooled liquid droplets are constrained by global satelliteobservations.The higher ECS estimates are directly linked to aweakened cloud-phase feedbackarising from a decreased cloud glaciation rate in a warmer climate.We point out the need forrealistic representations of the supercooled liquid fraction in mixed-phase clouds in GCMs,given the sensitivity of the ECS to the cloud-phase feedback.

Mixed-phase clouds, ubiquitous in Earth’satmosphere (1) at temperatures between0° and –40°C, strongly influence Earth’sradiationbudget (2–4). For a fixed amountof cloud water, spherical liquid droplets

tend to be smaller in size (5) and also to out-number ice crystals, because ice nuclei (IN) arerelatively scarce in Earth’s atmosphere in com-parison to cloud condensation nuclei (6). As aconsequence, clouds that consist of a higher frac-tion of liquid are optically thicker and hencemore reflective of sunlight. Here we refer to thefraction of supercooled liquid within a mixed-phase cloud at a particular isotherm as the su-percooled liquid fraction (SLF).It has recently been shown that SLFs are se-

verely underestimated on a global scale in a mul-titude of global climate models (GCMs) (7, 8).This arises from twomain causes. The first causeconcerns the challenge of representing the mi-croscopic nature of the variousmixed-phase cloud

processes (9) that cannot be resolved at the typicalspatial scales of GCMs. The Wegener-Bergeron-Findeisen (WBF) process for ice crystal growth isone such process that critically affects SLFs(8, 10–12). This process refers to the growth of icecrystals at the expense of surrounding super-cooled liquid droplets in a mixed-phase cloud asa consequence of the lower saturation vaporpressure over ice relative to liquid. An ambientvapor pressure in between the saturation vaporpressures over liquid and ice is both a necessaryand sufficient condition for the WBF process tooccur in mixed-phase clouds (5). GCMs typicallyparameterize the WBF process by assuminghomogeneous mixtures of supercooled liquiddroplets and ice crystals within a model gridbox(10, 11). However, in situ observations suggestthat suchmixtures rarely exist in nature. Instead,pockets composed purely of either supercooledliquid droplets or ice crystals that are orders ofmagnitude smaller than the volume of a typicalGCMgridbox are suggested to be the norm (13, 14).This implies that the WBF process may be tooefficient in GCMs and is thus potentially a culpritfor the underestimation of SLFs in GCMs. Themul-tiplemodes of ice nucleation (5), further complicatedby the uncertainty associated with the efficiency

224 8 APRIL 2016 • VOL 352 ISSUE 6282 sciencemag.org SCIENCE

1Department of Geology and Geophysics, Yale University, NewHaven, CT 06511, USA. 2Program for Climate Model Diagnosisand Intercomparison, Lawrence Livermore National Laboratory,Livermore, CA 94550, USA.*Corresponding author. E-mail: ivy.tan@yale.edu

RESEARCH | REPORTS

on A

pril

13, 2

016

Dow

nloa

ded

from

on

Apr

il 13

, 201

6D

ownl

oade

d fr

om o

n A

pril

13, 2

016

Dow

nloa

ded

from

on

Apr

il 13

, 201

6D

ownl

oade

d fr

om

of the various IN species (15), are additional fac-tors contributing to the uncertainties in SLFs.The second cause concerns the fact that mixed-

phase clouds are poorly constrained because ofthe difficulty of obtaining observations of theseclouds (16, 17). Although mixed-phase clouds areubiquitous in Earth’s mid- and high latitudes (1),in situ observations of these clouds are naturallylimited by sparse spatial and temporal coverage.Technical difficulties in distinguishing liquidand ice particles of varying sizes, as well as arte-facts associated with ice crystal shattering onprobes (18), further complicate this matter. Sat-ellite observations by NASA’s Cloud-AerosolLidar with Orthogonal Polarization (CALIOP)instrument (19) offer an attractive alternativeby providing global measurements of cloud ther-modynamic phases since 2006.To address the aforementioned issues, we

constrained cloud phase in version 5.1 of theNational Center for Atmospheric Research’sCommunity Atmosphere Model (CAM5.1) (20)by using 79 months of observations obtained byCALIOP. CAM5.1 is a state-of-the-art GCM, usedfor climate studies worldwide, and is among theGCMs that severely underestimate SLFs over theentire globe relative to satellite observations(7, 8). To constrain CAM5.1-simulated SLFs toagree with observations, we adopted a quasi–Monte Carlo sampling approach to select 256combinations of six cloud microphysical param-eters (table S1), among which we included theWBF process time scale for the growth of icecrystals and the fraction of atmospheric aerosolsactive as IN (12). The WBF process is renderedless efficient in each case by retarding the timescale at which the process occurs. The defaultCAM5.1 ice nucleation scheme is replaced withone that has been developed based on in situsurface and aircraft observations to prognosticallycalculate the ice nucleating-particle concentrationusing the concentration of large dust aerosols(21). Out of the 256 simulations, two parametercombinations (table S1) that yield root mean

square errors of SLF < 0.050 across all isothermswere implemented into the fully coupled versionof CAM5.1, CESM, version 1.0.6 (22), which in-cludes interactive full-depth ocean, sea-ice, andland components. To benchmark these satellite-constrained simulations (hereafter referred to asCALIOP-SLF1 and CALIOP-SLF2), fully coupledsimulations using the default model run withoutany modifications (henceforth referred to as con-trol), aswell as twomore caseswith unrealisticallyhigh and low SLFs meant to serve as the upperand lower bounds (henceforth referred to as high-SLF and low-SLF, respectively), were also run untilthe global net radiation budget at the top of theatmosphere was balanced with both present-dayand doubled CO2 concentrations, totaling 10 sim-ulations altogether (table S2 and fig. S1). The sim-ulated climate states are within realistic bounds(table S2). SLFs henceforth refer to those aver-aged over the past 50 years of simulation (the“mean state”) with present-day CO2 concentra-tions (the “initial state”).Themean extratropical (poleward of 30°) SLFs

of the five cases are compared against each otherin Fig. 1A, alongwith their ECS estimates in Fig. 1B.The correlation between SLF and ECS is evidentupon comparison of the two figures [Pearson cor-relation coefficient (R) = 0.98, P = 0.0025]. TheECS estimates monotonically increase with in-creasing initial-state SLF, exhibiting a rangeextending from3.9°C in low-SLF to 5.7°C in high-SLF. The ECS values of the other three casesexhibit a wide range of values in between thesetwo unrealistic extremes. The control case hasan ECS of 4.0°C, which is significantly lowerthan the satellite-constrained cases of 5.0° and5.3°C, for CALIOP-SLF1 and CALIOP-SLF2, re-spectively. This result suggests that GCMs thatunderestimate SLFs may also be severely under-estimating ECS. The upper bound of the obser-vationally constrained estimates is greater thanthat of 30 GCMs participating in the CoupledModel Intercomparison Project (CMIP, version 5)(23). It should be noted, however, that the ECS

estimates of these 30 GCMswere computed usingdifferent methods and thus may not be directlycomparable with the estimates presented here.This increase in the ECS estimates is directly

linked to a weakened negative cloud-phase feed-back that affects shortwave (SW)more stronglythan longwave (LW) radiation (2). In this feed-back, an initial doubling of CO2 concentrationscauses the entire troposphere to deepen. In re-sponse, isotherms throughout the tropospherewill shift upward in altitude relative to theirlocation in the initial state. This implies that atany fixed altitude, there is a greater likelihoodthat the SLF of any mixed-phase cloud occupyingthat altitude will be higher than that of any pre-existing cloud at the same altitude in the initialstate. Because mixed-phase clouds with higherSLFs are more reflective of SW radiation thanthose with lower SLFs, the enhanced reflectionof SW radiation back to space counteracts theCO2-induced warming. The cloud-phase feedbackbecomes less pronounced for mixed-phase cloudswith higher initial-state SLFs (24). To illustratethis effect in the extreme, the feedback effectivelyvanishes for mixed-phase clouds with SLFs ofunity, a scenario that is analogous to what occursinhigh-SLF (Fig. 2, A andC).Here, the replacementof ice with liquid after CO2 doubling only occursbetween temperatures extending from ~ –30°Cdown to –40°C, which is approximately the tem-perature threshold for homogeneous freezingof liquid droplets. Furthermore, its strength isweakened by the fact that less cloud condensateexists at colder temperatures. As the negativecloud-phase feedback weakens, it is less effectiveat compensating for other processes that reducecloud optical depth,whichmay include the dryingof cloud layers by convective mixing (25) andrapid cloud adjustments to CO2 (26). At the otherextreme is low-SLF, which exhibits the strongestcloud-phase feedback (Fig. 2, B and D). Here,the feedback occurs throughout the heteroge-neous freezing temperatures (0° to ~ –40°C). Al-though the cloud-phase feedback operates at alllatitudes, it is strongest in the extratropics. At highlatitudes, its effect on SW radiation is muted bythe polar night.The change in the gridbox-averaged liquid

water path (DLWP), which measures the changein vertically integrated cloud liquid water con-tent under global warming, monotonically de-creases with increasing initial-state SLF (Fig. 1C).This is consistent with a weakening of the cloud-phase feedback. Although DLWP is positive inlow-SLF, control, and CALIOP-SLF1 simulations,it eventually becomes negative in CALIOP-SLF2and high-SLF, when the cloud-phase feedback hasweakened to the point that other mechanismscause the overall negative DLWP.The mean extratropical net cloud optical

depth feedback (lt) quantifies the extent to whichchanges in the cloud optical depth amplify (lt > 0)or damp (lt < 0) the initial warming response toCO2 doubling. The cloud-phase feedback is aprimary but not the sole contributor to the cloudoptical depth feedback by virtue of the vastlydifferent optical depths of liquid and ice clouds

SCIENCE sciencemag.org 8 APRIL 2016 • VOL 352 ISSUE 6282 225

SLF

0.0

0.4

0.8

EC

S (

°C)

3.5

4.5

5.5

Low SLF

Control

CALIOP SLF1

CALIOP SLF2

High SLF

−∆

LWP

(kg

m−

2 )

−0.006

0.000

0.006

λτ

(W

m−

2 K

−1 )

0.4

0

0.4

0.8

Fig. 1. The link between cloud thermodynamic phase partitioning and ECS. Error bars represent the1s confidence interval. (A) The initial-state extratropical SLFs at the –10°C isotherm. (B) ECS estimatesin response to CO2 doubling. (C) Changes in global mean gridbox-average LWP. (D) Mean extratropicalnet cloud optical depth feedback, lt.

RESEARCH | REPORTS

(5). The monotonic increase in lt is also consist-ent with a monotonic weakening of the cloud-phase feedback with increasing SLF (Fig. 1D). ltis calculated based on the cloud radiative kernelmethod (27), using cloud fractions derived fromthe International Satellite Cloud ClimatologyProject satellite simulator (28, 29) implementedin CAM5.1. This method allows the total cloudfeedback to be decomposed into individual con-tributions from the changes in cloud optical depth(lt) and cloud-top pressure (i.e. height; lCTP) and

fraction (lAmt). Although the sum lCTP + lAmt

remains within 7% of that of the control across allcases, lt monotonically increases with increasingSLF over a wide range primarily because of dif-ferences in reflected SW radiation (Fig. 3A). Re-gressions of cloud and noncloud feedbacks onECS also strongly suggest that the cloud-phasefeedback is driving the spread in ECS throughits relation to lt (fig. S2).Global distributions of lt elucidate the strength

and locations over which the negative cloud-

phase feedback occurs. The negative cloud-phasefeedback is strongest in low-SLF, and the highlatitudes of both hemispheres show correspond-ing strongly negative lt values (Fig. 3B). Theweaker negative cloud-phase feedback in thecontrol case exhibits less negative lt values overthe high latitudes (Fig. 3C). Eventually, the neg-ative cloud-phase feedback is weakened to theextent that high-latitude lt values become in-creasingly positive through the manifestationof other warming mechanisms in this order:

226 8 APRIL 2016 • VOL 352 ISSUE 6282 sciencemag.org SCIENCE

Low-SLFHigh-SLF

Fig. 2.Weakening of the cloud-phase feedback. Pressure-latitude cross-sections of zonal mean-state changes in gridbox-averaged (A and B) cloud ice and(C and D) cloud liquid densities in [(A) and (C)] high-SLF and [(B) and (D)] low-SLF in response to CO2 doubling, exemplifying weakening of the cloud-phasefeedback. Isotherms in the present-day CO2 simulations are displayed as dashed lines; isotherms in the doubled CO2 simulations are displayed as solid lines.The quantity of ice throughout the mixed-phase clouds in low-SLFdecreases upon CO2 doubling, while the quantity of liquid increases.The feedback still occursin high-SLF, but it is substantially weakened, occurring only at the coldest isotherm.

Low

SLF

Con

trol

CAL

IOP

SLF1

CAL

IOP

SLF2

Hig

hSL

F

0.8

0.4

0.0

0.4

0.8

(W

m2 K

1 )

SW

LW

CTP+Amt

CALIOP-SLF1

Low-SLF

CALIOP-SLF2

Control

High-SLF

Wm-2K-1

-3

-2

-1

0

1

2

3

Fig. 3. Increase in the net cloud optical depth feedback parameter lt with increasing initial mean-state SLF. Error bars represent the 1s confidenceinterval. (A) Mean net extratropical lt, decomposed into its SW and LW components, alongside the sum of the mean net extratropical lCTP and lAmt. (B toF) Global distributions of lt.

RESEARCH | REPORTS

CALIOP-SLF1 (Fig. 3D), CALIOP-SLF2 (Fig. 3E),and high-SLF (Fig. 3F). It is worth noting that thecloud-phase feedback is particularly sensitive tohigher initial-state SLFs over the pristine SouthernOcean in the observationally constrained cases.Underestimates of SLFs relative to observationsin this region (30) lead to an artificially strongercloud-phase feedback that can ultimately lead toan underestimate of ECS. The negative lt valuesover the tropical Pacific are attributed to the thick-ening of high clouds (27).Global satellite observations of cloud thermo-

dynamic phases have enabled us to show thatunrealistically low SLFs common to a multitudeof GCMs lead to a cloud-phase feedback that istoo negative. This has important ramificationsfor ECS estimates. Should the low-SLF bias beeliminated inGCMs, themost likely range of ECSshould shift to higher values. It should be notedthat there are variations in the way in whichmicrophysical processes are parameterized inGCMs. Thus, the method by which SLFs can beconstrained in GCMs is not unique, and a repeatof this study in other GCMs may thus lead toother climate feedback responses that are notcaptured by the GCM used in this study. How-ever, regardless of how SLFs are constrained, theSLF will affect the magnitude of the cloud-phasefeedback and therefore GCMs’ estimates of ECS.Looking forward, continued improvements ofthe accuracy of various observational methods,better understanding of mixed-phase cloud pro-cesses, and their improved representation inGCMsare all factors that are critical to improving theaccuracy of ECS estimates.

REFERENCES AND NOTES

1. M. D. Shupe et al., Bull. Am. Meteorol. Soc. 89, 1549–1562 (2008).2. J. F. B. Mitchell, C. A. Senior, W. J. Ingram, Nature 341,

132–134 (1989).3. Z. Sun, K. P. Shine, Q. J. R. Meteorol. Soc. 120, 111–137 (1994).4. Z.-S. Li, H. Le Treut, Clim. Dyn. 7, 133–139 (1992).5. H. R. Pruppacher, J. D. Klett, Microphysics of Clouds and

Precipitation 1980 (Reidel, Dordrecht, Netherlands, 1978).6. B. J. Murray, D. O’Sullivan, J. D. Atkinson, M. E. Webb, Chem.

Soc. Rev. 41, 6519–6554 (2012).7. M. Komurcu et al., J. Geophys. Res. 119, 3372–3400 (2014).8. G. Cesana, D. E. Waliser, X. Jiang, J. L. Li, J. Geophys. Res. 120,

7871–7892 (2015).9. U. Lohmann, C. Hoose, Atmos. Chem. Phys. 9, 8917–8934 (2009).10. T. Storelvmo et al., Environ. Res. Lett. 3, 045001 (2008).11. J. Fan et al., J. Geophys. Res. 116, D00T07 (2011).12. I. Tan, T. Storelvmo, J. Atmos. Sci. 73, 709–728 (2016).13. A. Korolev, G. Isaac, S. Cober, J. W. Strapp, J. Hallett, Q. J. R.

Meteorol. Soc. 129, 39–65 (2003).14. P. Chylek, C. Borel, Geophys. Res. Lett. 31, L14104 (2004).15. J. D. Atkinson et al., Nature 498, 355–358 (2013).16. A. J. Illingworth et al., Bull. Am. Meteorol. Soc. 88, 883–898 (2007).17. H. Morrison et al., Nat. Geosci. 5, 11–17 (2012).18. A. Korolev, E. Emery, J. W. Strapp, S. G. Cober, G. A. Isaac,

J. Atmos. Ocean. Technol. 30, 2527–2553 (2013).19. D. M. Winker et al., J. Atmos. Ocean. Technol. 26, 2310–2323 (2009).20. R. B. Neale et al., Description of the NCAR Community

Atmosphere Model (CAM5.0). NCAR Tech. Note NCAR/TN-486+STR (2010).

21. P. J. DeMott et al., Atmos. Chem. Phys. 15, 393–409 (2015).22. J. W. Hurrell et al., Bull. Am. Meteorol. Soc. 94, 1339–1360 (2013).23. T. Mauritsen, B. Stevens, Nat. Geosci. 8, 346–351 (2015).24. Y. Tsushima et al., Clim. Dyn. 27, 113–126 (2006).25. S. C. Sherwood, S. Bony, J.-L. Dufresne, Nature 505, 37–42 (2014).26. T. Andrews, J. M. Gregory, P. Forster, M. J. Webb, Surv.

Geophys. 33, 619–635 (2012).27. M. D. Zelinka, S. A. Klein, D. L. Hartmann, J. Clim. 25,

3736–3754 (2012).

28. S. A. Klein, C. Jakob, Mon. Weather Rev. 127, 2514–2531(1999).

29. M. Webb, C. Senior, S. Bony, J.-J. Morcette, Clim. Dyn. 17,905–922 (2001).

30. D. T. McCoy, D. L. Hartmann, M. D. Zelinka, P. Ceppi,D. P. Grosvenor, J. Geophys. Res. 120, 9539–9554 (2015).

ACKNOWLEDGMENTS

This work was supported by NASA Headquarters under the NASAEarth and Space Science Fellowship Program, grant NNX14AL07H. Wealso acknowledge high-performance computing support fromYellowstone provided by National Center for Atmospheric Research’sComputational and Information Systems Laboratory, sponsored byNSF under grant 1352417. The effort of M.D.Z. was supported by theRegional and Global Climate Modeling Program of the Office ofScience at the U.S. Department of Energy (DOE) under grant

DE-SC0012580 and was performed under the auspices of DOE byLawrence Livermore National Laboratory under contract DE-AC52-07NA27344. CALIOP data are available online at https://eosweb.larc.nasa.gov/order-data. National Centers for Environmental Prediction–DOE Reanalysis 2 data are also available online at www.esrl.noaa.gov/psd/data/gridded/data.ncep.reanalysis2.pressure.html.

SUPPLEMENTARY MATERIALS

www.sciencemag.org/content/352/6282/224/suppl/DC1Materials and MethodsFigs. S1 and S2Tables S1 and S2References (31–37)

25 September 2015; accepted 3 March 201610.1126/science.aad5300

CANCER BIOLOGY

MYC regulates the antitumorimmune response throughCD47 and PD-L1Stephanie C. Casey,1 Ling Tong,1 Yulin Li,1 Rachel Do,1 Susanne Walz,2

Kelly N. Fitzgerald,1 Arvin M. Gouw,1 Virginie Baylot,1 Ines Gütgemann,1,3

Martin Eilers,2,4 Dean W. Felsher1*

The MYC oncogene codes for a transcription factor that is overexpressed in many humancancers. Here we show that MYC regulates the expression of two immune checkpoint proteinson the tumor cell surface: the innate immune regulator CD47 (cluster of differentiation 47)and the adaptive immune checkpoint PD-L1 (programmed death–ligand 1). Suppression ofMYC in mouse tumors and human tumor cells caused a reduction in the levels of CD47 andPD-L1 messenger RNA and protein. MYC was found to bind directly to the promoters of theCd47 and Pd-l1 genes. MYC inactivation in mouse tumors down-regulated CD47 and PD-L1expression and enhanced the antitumor immune response. In contrast, when MYC wasinactivated in tumors with enforced expression of CD47 or PD-L1, the immune response wassuppressed, and tumors continued to grow. Thus, MYC appears to initiate and maintaintumorigenesis, in part, through the modulation of immune regulatory molecules.

The transcription factor MYC regulates theexpression of a multitude of gene productsinvolved in cell proliferation, growth, self-renewal, differentiation, and apoptosis (1–4).The MYC gene is genetically activated and

overexpressed inmany human cancers (1–4), andthis overexpression has been causally linked totumorigenesis (5, 6). Studies involving indu-cible transgenic mouse models have shown thatgrowth of Myc-induced tumors is dependent oncontinuous expression of MYC (1–4, 7–10). Forexample, in the tetracycline-off mouse model(whereMyc expression can be turned off by theaddition of tetracycline or doxycycline), tumorsgrow onlywhenMyc is “on.”WhenMyc is turned“off,” tumors regress.

In mouse models, MYC inactivation results intumor regression through the induction of pro-liferative arrest and apoptosis (1–3, 7, 8, 10–12).We have demonstrated that complete tumor clear-ance that occurs after the inactivationof oncogenes,includingMyc, requires the recruitment of CD4+

T cells and the secretion of thrombospondin-1(13, 14).Hence, ahost-dependent immune responseis required for sustained tumor regression.However, the mechanism by which oncogene in-activationelicits this immune response is unknown.The host immune system generally serves as a

barrier against tumor formation (15). Activationof the immune response can contribute to tumorregression (13, 16, 17) through both adaptive andinnate immune effectors (18–20). Programmeddeath–ligand 1 (or PD-L1, also known as CD274and B7-H1) sends a critical “don’t findme” signalto the adaptive immune system (21–23), whereasCD47 (cluster of differentiation 47) sends a crit-ical “don’t eat me” signal to the innate immunesystem and acts as a regulator of the adaptiveimmune response (24, 25) (fig. S1A). These andsimilarmolecules are often overexpressedonhumantumors (22, 25). Therapeutic suppression of PD-L1

SCIENCE sciencemag.org 8 APRIL 2016 • VOL 352 ISSUE 6282 227

1Division of Oncology, Departments of Medicine and Pathology,Stanford University School of Medicine, Stanford, CA 94305,USA. 2Comprehensive Cancer Center Mainfranken, Core UnitBioinformatics, Biocenter, University of Würzburg, Am Hubland,97074 Würzburg, Germany. 3Institute of Pathology, UniversityHospital Bonn, 53127 Bonn, Germany. 4Theodor Boveri Institute,Biocenter, University of Würzburg, Am Hubland, 97074Würzburg, Germany.*Corresponding author. E-mail: dfelsher@stanford.edu

RESEARCH | REPORTS

DOI: 10.1126/science.aad5300, 224 (2016);352 Science

et al.Ivy Tanclimate sensitivityObservational constraints on mixed-phase clouds imply higher

This copy is for your personal, non-commercial use only.

clicking here.colleagues, clients, or customers by , you can order high-quality copies for yourIf you wish to distribute this article to others

  here.following the guidelines

can be obtained byPermission to republish or repurpose articles or portions of articles

  ): April 13, 2016 www.sciencemag.org (this information is current as of

The following resources related to this article are available online at

/content/352/6282/224.full.htmlversion of this article at:

including high-resolution figures, can be found in the onlineUpdated information and services,

/content/suppl/2016/04/07/352.6282.224.DC1.html can be found at: Supporting Online Material

/cgi/collection/geochem_physGeochemistry, Geophysics

subject collections:This article appears in the following

registered trademark of AAAS. is aScience2016 by the American Association for the Advancement of Science; all rights reserved. The title

CopyrightAmerican Association for the Advancement of Science, 1200 New York Avenue NW, Washington, DC 20005. (print ISSN 0036-8075; online ISSN 1095-9203) is published weekly, except the last week in December, by theScience

on A

pril

13, 2

016

Dow

nloa

ded

from